Interaction and transformation to deltas

Downstream analysis and statistics MaAsLin
1

Dear MaAsLin2 developers,

I am analyzing an intervention with two groups (A and B) and two time points (Basal and Final). My intention is to check for interaction; however, I believe I am specifying the model incorrectly because I cannot find any significant interaction. In fact, I have modified the data for a taxon to exaggerate the differences, but the results do not change. I am sending a reduced set of my data with 5 taxa and the modified final data for Blautia so that group A has higher abundance and group B has lower abundance
input_data.csv (4.5 KB)
input_metadata.csv (1.4 KB).

The codes used were as follows:

fit_data <- Maaslin2(input_data, 
                     input_metadata, 
                     output = 'demo_output',
                     random_effects = c("Subject"),
                     fixed_effects = c('TimeFinal', 'GroupB'),
                     standardize = F,
                     min_abundance = 0,
                     min_prevalence = 0.1,
                     analysis_method = "LM",         
                     normalization = "CLR", 
                     transform = "NONE",
                     max_significance = 0.15)

fit_data$results

On the other hand, can microbial data be transformed into deltas (difference between time points) for cross-sectional analysis with MaAsLin2?

Thank you very much for your work,
Bruno

2

Hi @Bruno - this has been discussed here.

All the best,
Himel

3

Hi @himel.mallick, thank you for your message. Before opening this topic, I had already read the messages, but I still don’t seem to get significant results with my data for the interaction variable, that is, ‘GroupB:TimeFinal’ does not appear in the ‘metadata’ column of the results, even when changing the data to increase the differences in Blautia abundance between the groups after the intervention. For this reason, I am requesting assistance with my case:

fit_data <- Maaslin2(input_data, 
                                input_metadata, 
                                output = 'demo_output',
                                random_effects = 'Subject',
                                fixed_effects = c('TimeFinal', 'GroupB', 'GroupB:TimeFinal'),
                                standardize = F,
                                min_abundance = 0,
                                min_prevalence = 0.1,
                                analysis_method = "LM",         
                                normalization = "CLR", 
                                transform = "NONE",
                                max_significance = 0.15)

fit_data$results

Additionally, I have tried to perform the analyses with the deltas, after subtracting the basal abundance from the final, but the model does not run, and I don’t understand if there is a specific aspect of the parameters to consider when the data is transformed into deltas. I would appreciate any example or comment on this matter.

Best regards,
Bruno

4

Hi @Bruno - I am confused when you say GroupB:TimeFinal’ does not appear in the ‘metadata’ column of the results, which tells me that this variable is not recognized by MaAsLin2. Would you be able to send your log file to confirm?

Thanks,
Himel

5

Sorry, @himel.mallick, here are the files again with the missing information:
input_metadata.csv (1.9 KB)
input_data.csv (4.5 KB)

6

Can you share the maaslin2 log file which should indicate what model MaAsLin2 is fitting?

Thanks,
Himel

7

I suppose you are referring to this information:
maaslin2.csv (3.5 KB)

8

@Bruno - thanks for sharing these files. Does the variable appear in the 'all_results.tsv' file but not in the 'significant_results.tsv' file? That would further validate that your model specification is correct, but the interaction effect is not significant.

For the analysis with deltas, you no longer need the subject random effect, time variable, and the interaction term. You can use a linear model, which can handle negative values, with no log transformation.

Hope this helps.

Thanks,
Himel

9

Hi @himel.mallick, in the all_results.tsv file, the interaction GroupB:TimeFinal does not appear anywhere. I don’t understand what is happening because, apparently, I have specified the model correctly. I would greatly appreciate your help in finding the error.

Regarding the model incorporating deltas, I understand that it would be sufficient to normalize the data using one of the methods because the model does not work without normalization.

Thank you very much for your patience,
Bruno